The present invention relates generally to a regular rule learning system, and more particularly, but not by way of limitation, to a system for learning regular structure over natural languages with variable lexical classes.
Mapping verbal usage to regular expressions have been considered. Conventional techniques proved that regular expressions extracted corpora can be learned and they are instrumental to a wide range of applications involving semantic processing. Such conventional techniques involve using of ontological categories.
Other conventional techniques rely on bags of words (i.e., a fixed number of lexical features) in order to predict the meaning of input content.
However, there is a technical problem with the conventional techniques that the ontological categories hinder the accuracy of the proposed method and the reliance on bag of words can limit the prediction of ambiguous terms.